Re: gdb_realpath: dealing with ./ and ../

[Mark Kettenis <mark.kettenis@xs4all.nl>]

> From: Aleksandar Ristovski <ARistovski@qnx.com>
> Date: Tue, 8 Jan 2008 11:11:33 -0500

Dropping in late, so excuse me if I'm saying things that already have
been said.  But I really want to make sure that any changes are no
going to hurt people that develop free software on POSIX systems where
they do have all source files available.

> If we can confirm for sure that "normalize_path" is safe, I think the idea
> is good (please read my comment about normalize_path here:
> http://sourceware.org/ml/gdb-patches/2008-01/msg00138.html)
> 
> Example
> DW_AT_name=../main.cc
> DW_AT_comp_dir=/foo/bar/obj
> The Directory Table:
>   ..
> The File Name Table:$
>   Entry>Dir>~~~~Time>~~~Size>~~~Name$
>   1>~~~~1>~~~~~~0>~~~~~~0>~~~~~~main.cc$
> 
> Now we get rid of the comp_dir information (this is what effectively
> happens):
> 
> NAME=main.cc
> DIR=/foo/bar
> 
> And there is no info about /foo/bar/obj.
> 
> 
> I would think that this is safe.

Unfortunately it isn't.  If /foo/bar/obj is a symlink to /bar/foo/obj,
then ../main.cc actually refers to /bar/foo/main.cc, and not
/foo/bar/main.cc.  So just "normalizing" names is defenitely unsafe.

The big question here is, whether a compiler will actually set
DW_AT_comp_dir to /foo/bar/obj in that case.  One can argue that the
compilation directory really is /bar/foo/obj in that case and that
DW_AT_comp_dir should be set to /bar/foo/obj.